You are on page 1of 1

LAB3: AI-based Android Mobile Application

for Skin Cancer Detection

Part 1: Skin Cancer Detection with EfficientNetV2

1.1 Download the dataset from Skin Cancer: Malignant vs. Benign (kaggle.com)

1.2 Define your custom CNN model based on the pretrained EfficientNetV2 model

(Classification with KerasCV ) to classify images into cancer and not cancer by modifying

this work.

1.3 Apply transfer learning and fine-tuning (Transfer learning & fine-tuning (keras.io)) of top

layers of EfficientNetv2’s feature extractor.

1.4 Add to your model these regularization methods weight decay (How to Use Weight Decay

to Reduce Overfitting of Neural Network in Keras - MachineLearningMastery.com), early

stopping (tf.keras.callbacks.EarlyStopping | TensorFlow v2.14.0) , and data augmentation

(Classification with KerasCV) to optimize your model’s performance.

1.5 Test and evaluate your model.

1.6 Save your custom CNN TensorFlow model and convert it to TFlite

(tf.lite.TFLiteConverter | TensorFlow Lite).

Part 2: AI-based Android Mobile Application

2.1 Follow the tutorial and add your TFlite Model into your Android mobile application.

2.2 Run the application, take a photo, or select an image from a folder, and get its prediction by

clicking on the button PREDICT.

1|Page

You might also like